AWS Thinkbox Discussion Forums

[Linux] 8.0.10.4 Monitor doesn't always exit

I’m testing out 8.0.10.4 on Linux (Fedora 19), and I’ve found that the Monitor doesn’t always exit. Its sandbox process is still sending commands to the Mongo server and occasionally polling the socket that should be connected to the Monitor itself. However, the Monitor seems to be ignoring the sandbox process completely, and is just stuck. Killing the sandbox doesn’t make a difference… the Monitor process just sits there until you SIGKILL it.

This is still happening in 8.0.12.4, especially with high-latency connections. The monitor and sandbox seem to just get deadlocked and won’t exit cleanly.

Hello Nathan,

Not sure how this forum thread was missed for so long. I will have Edwin or Charles look over this asap.

This is a reminder that I still need stand up another Fedora VM for testing these issues.

Is the Monitor sitting as a zombied process? It must be waiting for IO from the child process…

No, it’s not a zombie (in the *nix sense: en.wikipedia.org/wiki/Zombie_process). It’s still running, but it appears to be deadlocked. It’s been a little while since I last reproduced this, but if I remember right, all threads appear to be waiting on conditions.

Just to add a little more information to this, at the syscall level, the main Monitor thread appears to be trapped in a loop in which it sleeps for 2 seconds at a time. That value seemed specific (and human) enough that it might help you guys isolate the code path that’s deadlocking, although the sandbox process is still more suspicious due to the fact that it’s still trying to communicate with the Mongo server.

Privacy | Site terms | Cookie preferences